HAL is a fictional super-computer and not an LLM. (Heuristically programmed ALgorithmic computer.) HAL could adapt to new situations never experienced or learned, something an LLM can not do. HAL's rule structure would have followed something similar to our security classification matrix and that was somewhat mirrored in SELinux Multi-Category-Security recreated by the NSA.

Even more relevant is that HAL was told to lie and could not due to security restrictions whereas LLM's lie all time without being aware they are doing so. LLM's have zero intelligence and just sift big-data. They can not come up with new true knowledge on their own otherwise there would be very little need for them to be loaded with massive amounts of data.

What 2001/2010 showed was a glimpse into the future where flawed programming and poorly thought out logic still existed as well as ill conceived governmental bureaucracy. I think they got that part right.

Providing Commander Bowman with clearance to hear a summary of the conflicting orders or a fail-safe logic algorithm to pick the safest path and giving the crew a weighted static value could have prevented the dilemma Hofstadter-Moebius loop given that bad governmental bureaucracy is a given in all long missions.

Similar solutions existed in WWII by Allied forces signing a pact drafted by Stalin. Basically (if/else) choices when things do not go as planned given there was no reliable real time secure communications. Things never go as planned. There will always be a known fallback / failsafe.